Modern sensing technologies have enabled continuous monitoring of operating conditions in physical assets, providing significant amounts of data which helps to assess health conditions and degradation processes over time. This data is usually influenced by the interaction of a wide range of variables, making its understanding very challenging for standard parametric models in Prognostics Health and Management tasks. Novel data-driven approaches based on Deep Artificial Neural Networks have demonstrated to overcome such difficulties due to their versatility and capability in automatically interpret data and map it to a target health state. In this context, we propose a novel neural network model for the prognosis of remaining useful life (RUL) of an asset using multidimensional sequential sensor data: the Recurrent Capsule Network (RCN), based on the combination of Capsules Networks (CapsNet) and Recurrent Neural Networks (RNN). While CapsNets, a network composed by capsule-grouped neurons, has overcome the more traditional Convolutional Neural Networks (CNN) as features extractors to interpret spatial data, RNNs excel in sequential data analysis, demonstrating great performance in prognosis tasks. Within the proposed RCN model, spatial (sensorial) data is analyzed through capsules in a sequential manner with the RNN to assess the temporal evolution of the capsule data in order to obtain a more precise RUL prognostic. The proposed RCN model is applied to a public multi-sensor data set of turbofan engines to demonstrate its ability to predict remaining useful life, achieving state-of-the-art results.